Original Article

Proc IMechE Part G:

J Aerospace Engineering
Aerodynamic evaluation of an 2019, Vol. 233(13) 4980–4997
! IMechE 2019

unmanned aerial vehicle with Article reuse guidelines:

[Link]/journals-permissions

variable sweep and span DOI: 10.1177/0954410019836907

[Link]/home/pig

Cheng Gong and Bao-Feng Ma

Abstract
Morphing aircraft can achieve optimum performances at multiple flight conditions through large geometry deformations.
However, to obtain the optimum configurations, optimization design studies are required. A study on aerodynamic
optimization of a morphing aircraft was conducted to obtain corresponding optimal configurations at various flight
speeds. Firstly, an optimization framework being suitable to a morphing aircraft with larger deformations was established
by integrating existing codes, in which aerodynamic forces for the optimization are calculated by an Euler-based solver
and friction/form drag estimation code. The solver is based on a Cartesian method in which configurations are modeled
in terms of components of aircraft, hence large deformations of morphing aircraft can be performed during the opti-
mization. A surrogate-based model was employed for fitting aerodynamic forces, thus reducing computational cost in a
global optimization. A generic morphing aircraft with variable sweep and span was investigated at subsonic, transonic, and
supersonic conditions through the optimization process. The target of optimization is to obtain maximum lift-to-drag
ratios subject to lift, trim, and static stability constrains at each flight condition. The movement of center-of-gravity of the
aircraft was also considered in optimization. The results indicate that the center-of-gravity has an important effect on the
optimum configurations obtained, and the aerodynamic performance will be enhanced significantly if the center-of-gravity
is moved backward at transonic and supersonic cases. In the case of movable center-of-gravity, the optimum sweep
angles increase with the increase in flight speeds, and the optimum spans at transonic and supersonic speeds are smaller
than the subsonic case.

Keywords
Morphing aircraft, variable sweep, variable span, aerodynamic optimization

Date received: 12 September 2018; accepted: 14 February 2019

in which aerodynamic optimization is an important

Introduction tool in seeking optimum configurations for each
Morphing aircraft1 recently evoked a renewed interest flight condition.
in academic and industry fields in order to improve Aerodynamic optimization design of a morphing-
the multimission performances of aircraft, particu- wing aircraft is considerably different from a conven-
larly of unmanned aerial vehicles (UAVs). tional fixed-wing aircraft, and several new methods
Compared with traditional fixed-geometry aircraft, even philosophies are required to be considered and
morphing aircraft can effectively compromise the con- developed.3 For the fixed-wing aircraft, the optimiza-
flicts of performances at various flight conditions tion design is conducted around a baseline configur-
through significantly changing their shapes. The ation with relatively small changes in geometry, but
shape-morphing aircraft increase the performances for the morphing aircraft, larger deformation is
mainly through altering the shape of a wing, because required in optimization. In addition, although the
the wing is a main aerodynamic component that has a
key role in aerodynamic characteristics. Although
arbitrarily morphing the wing shape is the most Ministry-of-Education Key Laboratory of Fluid Mechanics, Beihang
ideal situation for morphing aircraft, a more efficient University, Beijing, China
and feasible manner is only morphing the fundamen-
Corresponding author:
tal parameters of a wing, such as airfoil profiles, sweep Bao-Feng Ma, Ministry-of-Education Key Laboratory of Fluid Mechanics,
angles, and span. Without doubt, developing morph- Beihang University, Beijing 100191, China.
ing aircraft rely on a joint multidisciplinary effort,2 Email: bf-ma@[Link]
Gong and Ma 4981

morphing aircraft can theoretically be tailored to an that attempted to extend nonlinear lift-line methods
arbitrary shape, the optimum configuration in an to the design of morphing aircraft.20 However, the use
optimization process is necessarily selected from the of high-fidelity computational fluid dynamics (CFD)
shape set obtained by the mechanisms driving the methods is not present for aerodynamic optimization
morphing.4 Existing studies on the aerodynamic opti- of morphing aircraft. Although CFD solvers have
mization of morphing aircraft consist of airfoil been applied to optimize a morphing trailing-edge
morphing4–8 and wing planform morphing.9–19 The wing,21 the geometric deformation in optimization is
wing planform has a number of parameters to relatively minor. Aerodynamic optimization for large-
change, e.g. sweep angle,9 span,10–14 wing twist,15–17 deformation morphing aircraft, such as variable
wing flexibility,18 among which the sweep and span sweep and variable span in a broad range of changing,
are the two key parameters, being able to more sig- still are required to be studied further.
nificantly influence the aerodynamic characteristics of Up until now, there are three types of approaches to
an aircraft at various flight conditions. Variable- CFD-based aerodynamic optimization. The first one is
sweep aircraft are not a new concept,1 and some the adjoint-based gradient method,22–26 the computa-
early aircraft with variable sweep-angle have been tional cost of which is independent in design variables,
developed and produced, such as the Soviet Union and therefore can deal with hundreds of variables.
MIG-23, the United States F-14, and the European However, the approach is sensitive to initial values,
Tornado. The variable-sweep design is motivated only being applicable to a local optimization with
by multimission requirements of aircraft in 1950s small deformation. Previous studies27 indicated that
and 1960s. At that time, it was found that the fixed- local optima commonly exist for a three-dimensional
wing aircraft would require larger wings for fulfilling optimization with a large design space, and therefore a
strict multimission objectives, and therefore would global optimization is recommended. The second one
be heavier than a morphing-wing design, as com- is the nongradient method (e.g. genetic algorithms),
mented by Weisshaar.1 Likewise, with disappearance which is applicable to a global optimization.
of these requirements subsequently, the variable- However, this method is extremely inefficient for opti-
sweep aircraft become unpopular progressively, mizing relatively complex shapes, and thus is mainly
because the variable-sweep design imposes consider- being employed in two-dimensional shape optimiza-
able penalties in weight for a single-mission aircraft tion.28,29 The third approach is the surrogate-based
due to relatively complex control mechanisms for approach,30–32 which can be used to a global optimiza-
changing geometries. tion, and will be more efficient than the gradient-based
Nevertheless, with growing interests for UAVs that method in the cases of fewer design variables (com-
are commonly required of multimission capacities monly less than 15 variables33).
since 1990s, the morphing aircraft concept raised a The present investigation is aimed to establish a
renewed interest. Therefore, the variable-sweep wing surrogate-based optimization framework by integrat-
becomes a promising technology for aircraft morph- ing the existing codes in which the optimum configur-
ing. Another promising morphing manner is the vari- ations for a generic morphing aircraft with variable
able span that is also able to greatly improve the sweep and span can be rapidly determined at any
performance of aircraft. The company Raython in flight condition using a high-fidelity CFD method.
the United States proposed an unswept variable- The varying parameters of the configuration, such as
span concept for multimission cruise missiles,10 and sweep angle, span, and angle of attack, will be chosen
the evaluation on the strategy indicated that the as design variables of optimization. The variable
design can increase 75% more loiter time at the end sweep is implemented in a rigidly-rotating manner,
of flight. Bae et al.11 calculated static and dynamic which is the simplest approach for a variable-sweep
aeroelasticity for a variable-span morphing UAV. aircraft. However, the rigidly-rotating variable sweep
Ajaj et al.13 and Woods and Friswell14 proposed raises difficulties in a CFD-based optimization,
new strategies for variable-span aircraft, and evalu- because a large variation in wing sweep can cause
ated the benefits in enhancements of aerodynamic topological variation of geometries, such as on the
characteristics and flight control. It can be anticipated junction of a wing and fuselage. This is difficult to
that the performance of morphing vehicles could be deal with using conventional grid deformation
improved further if the sweep and span are changed approaches. To the best of our knowledge, no previ-
together. An evaluation on dynamics and control ous studies were conducted on the optimization of
characteristics of a variable-sweep and variable-span rigidly variable-sweep aircraft using CFD-based
aircraft was conducted,19 but studies on aerodynamic methods up to date. Therefore, the first objective of
optimization were not involved. More importantly, the study is to develop a practical optimization
these existing optimization studies for morphing method being suitable to the morphing aircraft with
airfoils and wing planforms basically adopted low- large deformations, such as variable sweep and span,
fidelity methods, such as panel or vortex lattice by which the optimal configuration corresponding to
methods for calculating aerodynamic forces in opti- any flight condition of an aircraft can be obtained.
mization. In recent years, there still exist few studies The aerodynamic forces for optimization will be
4982 Proc IMechE Part G: J Aerospace Engineering 233(13)

computed by high-fidelity numerical simulations.

Although the current state-of-the-art for a CFD-
based optimization is based on Navier–Stokes equa-
tions, the Euler-based approach is still employed
extensively due to its higher computational effi-
ciency.23–25 A similar optimization process for large
deformation has been integrated in a previous study,34
but the specific details are completely different from
the present one. More importantly, based on the opti-
mization method, an aerodynamic optimization for a
wing-body-tail morphing aircraft with variable sweep
and span is carried out and the optimum configur-
ations of the aircraft were determined at subsonic,
transonic, and supersonic conditions. The effects of
associated design variables on aerodynamic character-
istics are also analyzed. Although other parameters,
e.g. wing twist, also have an effect on aerodynamic
characteristics of the aircraft, none of them belongs
to the morphing parameters in the present case.
Nonmorphing parameters can be optimized using
conventional optimization methods after the opti-
mum sweep and span has been determined. These
configuration parameters like a wing twist have been
studied extensively in previous studies, and there are
mature optimization methods.

Framework of optimization
The framework of the optimization process is shown
in Figure 1, which consists of several modules. The
whole process of optimization is integrated using a
platform program with a graphic interface coded by
MATLABTM language. The optimization process is
based on a surrogate model (response surface) and a
high-fidelity Euler-based solver, combined with a fric- Figure 1. Flowchart of response surface construction.
tion/form drag estimation code. A genetic algorithm
in conjunction with a gradient algorithm is used to
optimize the response surface. The surrogate-based performed to obtain reasonable distributions of
methods were extensively applied on aerodynamic design variables.
optimization in the aerospace community.30,31 The In the optimization framework, there are several
current optimization study for a morphing aircraft associated numerical tools used in the optimization
involves large-scale variable-sweep and span, and process. These numerical tools are listed below, con-
therefore it is difficult to guarantee that there is neces- sisting of geometric parameterization, CFD solver
sarily one single local minimum in the design space. (Euler solver), friction/form drag estimation, surrogate
Therefore, a global optimization algorithm (e.g. gen- models, and optimization algorithms. At the end of
etic algorithm) is necessarily employed rather than a this section, the CFD solver and code of friction/
local gradient algorithm only. However, the global form drag estimation is validated using standard
algorithm based on CFD is extremely time-consuming experimental models and associated experimental data.
and affordable, and the existing studies for the CFD-
based optimization basically adopt local algorithms.
To avoid the high computational cost of a global opti-
Geometric parameterization
mization, a surrogate model based on a response sur- The geometric parameterization is done using an open
face is employed, and the global optimization is source code OpenVSP that is a conceptual design tool
conducted on the surrogate model. In order to further for aircraft.35 The model in OpenVSP is divided into
promote the optimization speed, a local gradient algo- various components, such as a wing, body, horizontal
rithm is also employed with initial values that come tail, and vertical tail, which are parameterized separ-
from the final results of the global optimization. In the ately. The current investigation only changes the wing
process of constructing the response surfaces, design- shape, which is parameterized by planform param-
of-experiments with a full factor analysis is firstly eters, e.g. span and sweep. The wing parameters can
Gong and Ma 4983

be modified by hooking API functions of OpenVSP of volume meshes are extremely important in the opti-
that can be accessed through a C-like language script mization of morphing aircraft with large deform-
(Angelscript). By coding the script program, the ations, which guarantees that the large morphing,
design variables of the wing can be modified. The such as variable sweep and span, can be implemented
model can be natively created in OpenVSP, or made in the optimization, particularly for the aircraft with
using other modeling software, ultimately being rigidly-rotating variable sweep.
output into the OpenVSP and re-parameterized.36 A second-order van Leer scheme was selected for
the flux computation, and a van Leer limiter was used
in a supersonic computation to suppress oscillation of
CFD solver shockwave. The temporal scheme is a fifth-stage
In the optimization process, the CFD solver is called Runge–Kutta scheme. The wall is set as a slip bound-
repeatedly to obtain aerodynamic force coefficients. ary condition, and the outer boundary of a domain is
The CFD solver is an inviscid Euler solver Ansys set as a far-field condition.
Cart3d (a commercial version of the Cart3d developed
by Aftosmis37), which is based on octree-based
Friction/Form drag estimation
Cartesian grids and a finite volume method. The ori-
ginal version of Cart3D has been extensively Since the CFD solver is based on inviscid Euler equa-
employed in the conceptual design of aircraft.38,39 tions, the friction/form drag will be unable to be
Compared with viscous CFD solvers based on obtained. Herein, a friction/form drag estimation pro-
Navier–Stokes equations, an Euler solver is much gram was employed to calculate the additional drag.
more efficient, thus being extensively applied in aero- The estimation of the friction/form drag adopts the
dynamic optimizations. In aircraft design, various method of Gur et al.40 that can provide an estimate of
fidelity models are all required for aerodynamic opti- laminar and turbulent skin friction and form drag for
mization to satisfy requirements in different design each component of an aircraft. The estimation
phases. The choice for computational tools depends method has been successfully applied to the high-fide-
on a tradeoff between efficiency and accuracy. The lity shape optimization of aircraft in previous stu-
Cartesian volume meshes required by Cart3D can be dies.25 The friction/form drag for fuselage-like
generated automatically with watertight surface components is estimated separately from the wing-
meshes of triangularization; the surface mesh for like components, and the total drag is the drag
each component of aircraft can be created individu- summation of all components. The formula for the
ally, and an intersection program in the Cart3D tools estimation of each component is illustrated in
can be employed to carry out a Boolean operation for equation (1), in which Swet and Sref are the wetted
these surface meshes, obtaining the watertight surface and reference area; CF is a friction coefficient and
mesh. One example of the mesh intersection is illu- FF is a form factor. CF uses standard flat-plate skin-
strated in Figure 2, in which additional triangular friction formulas, and the compressibility effects are
grid elements will be created on the junction between included by using the Eckert reference temperature
two components, none of which is skew. method for a laminar flow and the van Driest II for-
Subsequently, the volume meshes can be generated mula for turbulent flow. A composite formula is uti-
smoothly based on the watertight surface mesh. The lized to consider the situation of partially laminar and
fully intersected watertight surface meshes can also be turbulent flow. Form factors FF is employed to esti-
obtained using OpenVSP. In the present study, the mate the effect of thickness on drag, and associated
intersection tool of Cart3D was used. The intersection formulas for a body of revolution and wing are shown
operation of surface meshes and automatic generation in equations (2) and (3), in which d and l are the

Figure 2. Surface meshes before (a) and after (b) intersection operation.
4984 Proc IMechE Part G: J Aerospace Engineering 233(13)

diameter and length of a body, and t and c are the optimization problem. The advantage of the genetic
thickness and chord length of a wing. Equation (1) is algorithm is being able to identify a global optimum
valid from subsonic to moderate supersonic speeds solution, but the accuracy is not very good. Therefore,
(about Mach 3) a better choice is that a gradient algorithm is
employed to further optimize the response surface
Swet with the solution of the preceding global optimization
CD,f ¼ CF FF ð1Þ
Sref as initial values. By doing this, the gradient algorithm
can search the optimum solution along a gradient dir-
 1:5  3
d d ection. The global genetic algorithm is configured
FFbody ¼ 1:0 þ 1:5 þ 50 ð2Þ with a population size of 100, one-point cross-over
l l
and exchange mutation operator. A sequential quad-
t  t 4 ratic programming method was used for the local gra-
FFwing ¼ 1:0 þ 2:7 þ 100 ð3Þ dient-based optimization. The upper and lower
c c
bounds for the gradient-based optimization are the
Note that the location of laminar-to-turbulent same with the global process.
transition is provided as a given parameter in the fric-
tion/form drag estimate. A simple approach here is
used to determine the transition location for cases
Validation of numerical methods
with natural transition, as done in Gur et al.,40 in The Cart3D code as a mature inviscid Euler solver has
which two empirical curves for transition Reynolds been verified and validated extensively.37 Herein to
number based on streamwise characteristic scales are validate the reliability of the hybrid method by com-
presented, and therefore the transition locations can bining the Euler solver and friction/form drag estima-
be obtained directly by a table lookup. By this tion, comparisons were made with two standard
method, boundary layer parameters such as Re- are models at transonic and supersonic speeds, in which
not required, as in general transition models. reliable experimental data are available. The asso-
ciated results are shown in Figures 3 to 5. Note that
not all available experimental data could be suitable
Surrogate models for validating CFD results, and the systematic errors
A response surface model is employed as a surrogate and uncertainties in experiments should be strictly
model, which is computationally efficient and robust controlled. The experimental data herein satisfy this
and is applied extensively in aircraft optimization requirement. The following computations are all
design. A third-order response surface model was based on a half model.
used, as listed below The aircraft model for validating the transonic
computation is the DLR-F6 model that can be
y~ ¼ 0 þ 1 x1 þ   þ M xM þ Mþ1 x21 þ Mþ2 x22 þ   found in the drag prediction workshop website.41 A
grid convergence testing is first carried out to study
þ 2M x22M þ 2Mþ1 x31 þ 2Mþ2 x32 þ 
X the effects of grid discretization, as shown in
þ 3Mþ1 x33M þ ij xi xj Figure 3(a). The lift and drag coefficients are approxi-
i6¼j mately unchanged as the grid amount is more than
0.6 million. Figure 3(b) shows the comparison with
By estimating the errors of response surfaces, it is experimental data, in which the numerical results
found that the third-order response surface is sufficient overall agree well with the experiments. The transonic
as a surrogate model to represent the aerodynamic case is generally more difficult to simulate due to the
force coefficients. Design-of-experiments based on a mixed flows with subsonic and local supersonic flows.
full-factorial analysis is utilized to construct the distri- The good consistency herein is attributed to the lower
bution of sample points in the response surface. angle of attacks (AOAs) (less than 2 ). For our simu-
The response surface employed was validated for lation on the morphing aircraft, the cruise AOAs will
accuracy, and the associated validation was presented be restricted below 3 for all flight conditions. At
below in combination with the specific results of lower AOAs, no significant flow separation exists
optimization. around the model, and therefore the inviscid simula-
tion can have a rather good performance in predicting
inviscid drag. The associated pressure distributions at
Optimization algorithm
various spanwise locations (y/s) also show a good
The optimization algorithms for the response surface agreement with the experimental results (Figure 4),
are a global method (genetic algorithm) in conjunc- indicating that the flow fields were simulated cor-
tion with a local method (gradient algorithm). The rectly. In addition, no experimental data for skin fric-
genetic algorithm is inspired by the biological evolu- tion strain could be available for comparison.
tion to obtain the solutions of optimization problems, A supersonic aircraft model42 was employed to val-
and is extensively applied to solving a global idate the supersonic computation, which is the only
Gong and Ma 4985

(a) (b)

Figure 3. Numerical validation for DLR-F6 model at Ma ¼ 0.75: (a) grid convergence; (b) experimental comparison.

(a) (b)

(c) (d)

Figure 4. Pressure distributions at Ma ¼ 0.75 and CL ¼ 0.5: (a) y/s ¼ 0.15; (b) y/s ¼ 0.331; (c) y/s ¼ 0.411; (d) y/s ¼ 0.638.

available experimental model with reliable supersonic and experimental results agree well with each other.
aerodynamic data. Grid convergence testing and Note that the AOA for the maximum lift coefficient is
experimental comparison are shown in Figure 5. 31.5 at the supersonic case; at such high AOAs, the
The supersonic computation is more prone to con- computational results are still extremely close to the
verge with increasing grid cells, in which the results experimental values. The reason being that the wave
are basically unchanged as grid cells are more than 0.1 drag accounts for most of the total drag in the super-
million. The comparison shows that the numerical sonic case, which is able to be captured correctly by
4986 Proc IMechE Part G: J Aerospace Engineering 233(13)

(a) (b)

Figure 5. Numerical validation for a supersonic model at Ma ¼ 2.5: (a) grid convergence; (b) experimental comparison.

an inviscid solver. In addition, fully supersonic flows airfoils of a maximum thickness of 6% chord
are basically linear at a condition of small disturbance length. The horizontal tails are located 0.07 m lower
(e.g. low AOAs), it is easier to simulate. Therefore, below the horizontal symmetry plane of the fuselage.
although no experimental data for pressure and skin The deflection angle ’ of the horizontal tails will be
friction strain could be available for comparison, the changed for trim during optimizaiton. The rear of the
numerical results should be reliable. fuselage is streamlined to avoid any flow separation.
In addition, the above results were computed and
obtained as an AOA sweep for the transonic and
Movement of center-of-gravity
supersonic cases (the AOA range is the same with
the experiment), but were presented with a polar The center-of-gravity (CG) of the morphing aircraft is
curve of CD against CL. The reason for this represen- able to move along the longitudinal direction of the
tation is that the AOA is a design parameter in the fuselage as the CG of the wings varies owing to chan-
optimization design and will varies subject to a fixed ging of span and sweep angle, as shown in Figure 7.
lift constraint. In such a case, the accuracies of drag Movement of CG can influence the pitching moment
coefficients at a fixed lift coefficient are more import- and static stability of the aircraft, so it is necessary to
ant. This polar curve is also a conventional represen- quantify the movement amount of CG. For doing
tation in validating numerical results with this, the CG of the whole aircraft (CGt) is decomposed
experimental data (see Drag Prediction Workshops into the wing CG (CGw) and the CG of the fuselage
for details41). and tails (CGb). The CGb is normally fixed, but the
CGw can vary with the wing morphing. The wing
planform is rectangular with equal cross-sections,
Optimization setup and therefore the CG of each wing is located at the
midpoint of the wing. Through a simple derivation,
Geometry of morphing aircraft the distance from the CGw to the wing pivot Xw can be
The current morphing aircraft with variable sweep obtained by the following
and span is designed as a small unmanned aerial vehi-
cle, as displayed in Figure 6. The aircraft is a generic 1
Xw ¼ ðs sin l þ c cos lÞ
wing–body–tail configuration, in which the wing is a 2
uniform straight wing with variable sweep angle l,
and span s. The sweep angle can be altered between where l is the sweep angle, c is the chord length, and s
0 –80 by a manner of rigid rotation around the pivot is the span. The chord length is used as a reference for
located at the apex of the wing root. The wing span is defining the position of the CG. Since the wing is a
allowed to expand and retract in a range of 1.5–2.5 m. straight wing with equal cross-sections, the chord
The airfoil for the wing is the NACA 0012 with a length also equals a mean aerodynamic chord. For
maximum thickness of 12% chord length. The wings other irregular wing shapes, similar formulas also
are located 0.07 m higher above the horizontal sym- can be derived simply with more complicated forms.
metry plane of the fuselage. The fuselage is an axisym- For the small UAV, the weight of the wings is
metric ogive-cylinder body, and the vertical and estimated at 10% of the total weight. The coordinate
horizontal tails are all swept with the NACA 0006 origin is set at the nose tip of the fuselage, it is
Gong and Ma 4987

Figure 6. Geometry of a morphing aircraft with variable sweep and span (in meter).

variable-sweep (e.g. F-14 Tomcat) or supersonic

cruise (e.g. Concorde) aircraft are commonly required
for being able to artificially move CG through pump-
ing fuel forward and backward in equilibrium fuel
tanks, which can be easily implemented by automatic
control of computers.43 In the present case, artificially
adjusting the CGb would result in significant move-
ment of the CGt, and the effects of the CGb movement
on the stability and lift-to-drag ratios will be demon-
strated in the ‘‘Conclusions’’ section. In addition, no
any flexibility of wings was considered in the investi-
gation, which is worth studying in the future.44

Figure 7. Movement of center-of-gravity owing to altering Computational grids

sweep angle and span.
The surface grids were generated using a triangular-
ization algorithm based on parameterized models in
obtained by CGw ¼ 1:9 þ Xw and CGt ¼ 0:9CGb þ OpenVSP, and the volume grids were created based
0:1CGw . The CGt is the CG of the aircraft and is on the surface grids by the Cartesian grid generator
also the reference point of aerodynamic moment in Cubes in the Cart3D tool set. Grid cells inside the
optimization. model will be removed, and the volume cells inter-
Since the wing weight accounts for a small propor- sected with the surface grids will be cut away by the
tion in the total weight, the CGt movement caused by triangular surface meshes. Therefore, the cut-cells
the variation of the CGw is limited. However, the aero- near the surface are irregularly polyhedral, but not
dynamic center of the aircraft can be significantly necessarily being hexahedral. Entire computational
moved backward with increasing sweep angle, par- domains are hexahedral for all cases, and therefore
ticularly at supersonic speeds. This will cause the air- the specific domain sizes are different in various
craft to be strongly stable, thus leading to larger trim flight conditions. One example of the surface and
drag and even being difficult to control in pitch. volume grids is illustrated in Figure 8(a).
Therefore, static stability margin is generally designed Since there is no sideslip angle, all simulations are
at 0.05–0.15c for maintaining a good stability per- based on a half model in which a symmetry condition
formance. In order to fulfill this requirement, was imposed at the symmetry plane. Before a formal
4988 Proc IMechE Part G: J Aerospace Engineering 233(13)

(a) (b)

(c)

Figure 8. Grid convergence testing at Ma1 ¼ 0.85: (a) computational grids; (b) CL convergence; (c) CD convergence.

optimization, a grid convergence study for the morph-

ing aircraft was conducted at transonic and super- Table 1. Three flight conditions.
sonic conditions to determine a suitable grid Flight conditions Ma1 H (m) W (kg) CL
quantity for the optimization. The results indicated
that the supersonic case can converge more quickly 1 0.5 20 1200 0.29
than the transonic one with increasing grid amount, 2 0.85 5000 1200 0.19
which is consistent with the results in Figures 3 to 5. 3 2.0 300 1200 0.019
The lift and drag coefficients for one configuration
with l ¼ 19, s ¼ 1.9 m, and  ¼ 2.2 at the transonic
condition (Ma ¼ 0.85) are shown in Figure 8(b) and
(c), in which CL retains a minor variation after the
Optimization problem formulation
grid cells are more than 0.3 million; while CD shows 1. Objective function
convergence until 1.3 million cells. Since large amount
of CFD computation during the optimization will be The present aerodynamic optimization seeks to
performed for each case, the grid amount for various maximize lift-to-drag ratios subject to the constraints
flight conditions is optimized in terms of the results of on lift coefficients, static stability margin and trim at
the grid convergence by comprising computational subsonic, transonic, and supersonic flight conditions,
efficiency and accuracy. The specific grid cells for which is equivalent to minimizing drag for a fixed lift
the optimization are 1.37 (subsonic), 1.33 (transonic), target. The specific flight conditions are given in
and 1.22 (supersonic) million cells. Table 1, in which different flight speeds correspond
Gong and Ma 4989

to different flight altitudes. Although three flight various flight conditions, hence the lift constraint is
conditions were conducted for optimization, they constant and associated lift coefficients vary with
are optimized separately and each case is a single- flight conditions. The static stability in pitch is also
point optimization. Note that conventional aero- an important parameter, which should be con-
dynamic optimization is commonly based on a refer- strained. The longitudinal stability is enforced to
ence configuration, but the variable sweep and span remain 5–15% of the chord length
aircraft will tailor their configurations to an optimum
one at each flight condition, and therefore no refer- 0:054ðxnp  xcg Þ=c ¼ @CM =@CL 40:15
ence configurations exist.
where the xnp denotes the neutral point and the xcg is
2. Design variables the CG (herein CGt). The partial derivative @CM =@CL
is evaluated using a second-order centered difference
The design variables consist of sweep angle l, span scheme ðCM0 þ  CM0  Þ=ðCL 0 þ  CL 0  Þ, in
s, deflection angle of horizontal tails ’, angle of attack which the step  relies on a compromise between
, and center-of-gravity (CGb) as shown in Table 2. roundoff and truncation errors. In our case, the pitch-
The l and s belong to morphing parameters that are ing moment and lift coefficients are basically linear
quantities of most interest, and ’ is used to obtain owing to the lower a considered, and therefore the
trim constraints. Although there are only five design @CM =@CL is insensitive to . The effect of  on
variables in our study, each variable will be discretized the calculation of the partial derivative has been
into a series of sampling points to construct response tested, and the results almost remain unchanged as
surfaces. Owing to the wider variable bounds, too few  varies between 0.001 and 0.25 , and here the
sampling points are unable to capture the details of step of 0.1 was selected for formal computation.
the response surfaces. In order to fulfill the require- For the morphing aircraft with variable sweep, the
ments of accuracy, new data points are necessarily upper bound of the longitudinal stability can restrict
added in the sampling process. Eventually, a total of the maximum of sweep angles in optimization, thus
720, 2400, and 1152 samples are calculated for sub- further limiting an increase of lift-to-drag ratios, espe-
sonic, transonic, and supersonic cases (see cially in a supersonic case. Therefore, the variable-
‘‘Conclusions’’ section for details). Additionally, the sweep aircraft are commonly required to adjust the
static stability constraint is computed using a second- CG at various conditions in order to obtain better
order centered differencing scheme, and therefore performance (see section ‘‘Transonic condition
each sample requires three data at adjacent AOAs (Ma1 ¼ 0.85, H ¼ 5000 m)’’). Herein, the effects of
to calculating the differencing, and therefore CFD cal- the CG (CGt) on the optimization will be studied
culations are invoked for 2160, 7200, and 3456 times through considering the CGb as a design variable in
for three flight speeds. The design space is large optimization.
enough for the optimization with geometries of large
deformation.
Accuracy of surrogate models
3. Constraints Accuracies of the response surface models employed
was estimated by comparing the prediction values of
For any optimization design, constraint conditions surrogate models with direct CFD results, in which 19
are required to obtain practical optimization results. additional sample data were selected randomly by the
Herein, it is assumed that the weight of the morphing Latin hypercube approach and none of them belongs
aircraft is unchanged with a constant of 1200 kg at to the samples for constructing the surrogate models.
An example for the comparison is illustrated in
Figure 9 where the solid lines represent predictive
values, while the circle points represent calculated
Table 2. Optimization problem setup. values of CFD. It can be seen that the predictive
Objective function/Variables/Constraints
values agree well with the CFD ones.
To quantify the error level of the fitting, the
Maximize CL/CD Lift-to-drag ratio R-squared (R2) index was calculated for the response
With respect to  Angle of attack surfaces of CL/CD, CL, and CM at three flight condi-
 Sweep angle tions, as listed in Table 3. R2 is computed by the sum
s Span of squares of the distances between predictive values
’ Deflection angle of and calculated data. This sum-of-squares value is
horizontal tails called Sreg. To turn R2 into a fraction, the results
Subject to CL Lift coefficient constraint are normalized to the sum of the square of the dis-
@CM =@CL Static stability tances of the points from a horizontal line through the
mean of all values, and this value is called Stot.
CM Pitching moment
Therefore, R2 ¼ 1.0  (Sres/Stot). R2 is a fraction
4990 Proc IMechE Part G: J Aerospace Engineering 233(13)

(a) (b)

Figure 9. Comparison of direct CFD values and prediction values of surrogate model: (a) CL; (b) CL/CD.

Table 3. Accuracy of surrogate models. Table 4. Bounds of design variables at Ma1 ¼ 0.5.

R2 Ma1 ¼ 0.5 Ma1 ¼ 0.85 Ma1 ¼ 2.0 Design variables  (deg)  (deg) s (m) f (deg) CGb (m)

CL 1.000 0.997 0.983 Lower bound 0.0 0.0 1.5 0.0 2.2
CL/CD 0.991 0.965 0.975 Upper bound 3.0 30.0 2.5 5.0 2.4
CM 0.924 0.998 0.957

characteristics, and a low-sweep wing can have a

between 0.0 and 1.0, and has no units. Higher values better performance at subsonic speeds by previous
indicate that the model fits the data better. When R2 design experience. Therefore, the variation range of
equals 0.0, the best-fit curve is no better than a hori- sweep angle of 0–30 is reasonable for searching the
zontal line going through the mean of all calculated optimum configuration in the subsonic case, and a
values. When R2 ¼ 1.0, all points lie exactly on the pre-optimization process with a coarser variable inter-
curve with no scatter. As a general rule, the R2 val was conducted to validate the reliability of choos-
values of more than 0.9 indicate that accuracy of ing the range. The pre-optimization also indicated
response surfaces is acceptable. Table 3 shows that that the optimum lift-to-drag ratio is located at the
the R2 values in all cases are more than 0.96, indicat- range of angle of attack of 0–3 . The computational
ing the response surfaces have a good accuracy. domain is a hexahedron of 20 times body length
upstream of the nose tip, 20 times body length down-
stream of the bottom end of the fuselage, and 20 times
Optimization results body length from a symmetry plane to the side bound-
The optimization results for subsonic, transonic, and ary. Total grid cells are around 1.37 million that can
supersonic conditions are presented in the section. vary slightly with re-meshing the volume meshes
These three conditions are optimized independently during the optimization.
and the optimum configuration for each condition is The sampling points of each design variable are
achieved. The effects of the design variables on the generated using a full factor design, and totally 720
optimization objective are also analyzed. sampling points are employed for constructing
response surface models. Based on the sampling
points, the CFD solver and friction/form drag estima-
Subsonic condition (Ma1 ¼ 0.5, H ¼ 20 m) tion code are invoked for 2160 times to calculate aero-
In this subsection, the optimum configuration for the dynamic coefficients and static stability, eventually
morphing aircraft flying at a subsonic and low- obtaining the response surfaces. The global optimiza-
altitude condition (Ma1 ¼ 0.5 and H ¼ 20 m) are pre- tion algorithm (genetic algorithm) was first carried
sented. The searching range for design variables is out on the response surface, and the optimum param-
shown in Table 4, in which the span (s ¼ 1.5–2.5 m) eters obtained were taken as initial values for the local
takes values in the entire design space, but the sweep gradient optimization. The results from the two opti-
angle is restricted in the range of 0–30 . The sweep mizations are shown in Table 5, and Figure 10 shows
angle has a significant effect on aerodynamic the optimal configuration and associated pressure
Gong and Ma 4991

Table 5. Optimization results at Ma1 ¼ 0.5.

Algorithm  (deg)  (deg) s (m) f (deg) CGb CL CL / CD CM @CM =@CL

Genetic 2.553 14.459 2.344 2.929 2.3 0.290 11.682 0.000 0.126
Gradient 2.643 15.600 2.291 3.067 2.3 0.290 11.715 0.000 0.146

Figure 10. Optimization results at Ma1 ¼ 0.5 and H ¼ 20 m: (a) optimum configuration and (b) associated flow field.

Figure 11. Response surfaces at Ma1 ¼ 0.5 and H ¼ 20 m: (a) CL/CD vs.  and s; (b) CL/CD vs.  and ; (c) CL/CD vs. s and ; (d) CL/CD
vs. f and .
4992 Proc IMechE Part G: J Aerospace Engineering 233(13)

Table 6. Bounds of design variables at Ma1 ¼ 0.85.

Design variables  (deg)  (deg) s (m) f (deg) CGb (m)

Lower bound 0.0 5.0 1.8 0.0 2.2

Upper bound 2.0 55.0 2.5 6.0 3.1

Table 7. Optimization results at Ma1 ¼ 0.85.

Algorithm  (deg)  (deg) s (m) f (deg) CGb (m) CL CL / CD CM @CM =@CL

Genetic 1.984 46.302 1.811 1.317 2.8 0.19 6.952 0.000 0.074
Gradient 2.000 46.372 1.803 1.300 2.8 0.19 6.962 0.000 0.070

(a) (b)

Figure 12. Effects of CG movement at Ma1 ¼ 0.85: (a) CL/CD vs. CGb; (b)  vs. CGb.

Figure 13. Optimization results at Ma1 ¼ 0.85 and H ¼ 5000 m: (a) optimum configuration and (b) associated flow field.

contour. The optimum values from the gradient opti- sweep angle of 15.6 and span of 2.3 m conform to
mization are similar with the global optimization, the characteristics of subsonic configurations.
indicating that the genetic algorithm has approached The optimum CGb is located at the position of 2.3 m
the optimal solution. The optimum values with a to the nose tip, which is also chosen as a reference
Gong and Ma 4993

Figure 14. Response surface at Ma1 ¼ 0.85 and H ¼ 5000 m: (a) CL/CD vs.  and s; (b) CL/CD vs.  and ; (c) CL/CD vs. s and ;
(d) CL/CD vs. f and .

Table 8. Bounds of design variables at Ma1 ¼ 2.0.

Design variables  (deg)  (deg) s (m) f (deg) CGb

Lower bound 0.0 50.0 1.8 0.0 2.85

Upper bound 1.5 80.0 2.5 6.0 3.2

location for comparison with the transonic and super- around 13 (Figure 11(a)), whereas the CL/CD basic-
sonic cases. ally monotonically increase up to the upper bounds
The response surface can also be used to analyze the (Figure 11(a) to (d)) with increasing the span and
effects of design variables on lift-to-drag ratios, but the angle of attack. However, with the deflection angle
total response surface with three variables is difficult to of horizontal tails ’ increasing, the CL/CD decreases
display by graphs, hence the sub-surfaces are illu- owing to larger trim drag (Figure 11(d)).
strated in Figure 11, in which the lift-to-drag ratio
against two design variables are presented by keeping
Transonic condition (Ma1 ¼ 0.85, H ¼ 5000 m)
the other variables being constants (optimum values).
By these results, the effects of the design variables on In this subsection, the optimum configuration for
the lift-to-drag ratio can be revealed. The CL/CD morphing aircraft at a transonic and middle-altitude
apparently exhibits a maximum at a sweep angle of condition (Ma1 ¼ 0.85 and H ¼ 5000 m) are
4994 Proc IMechE Part G: J Aerospace Engineering 233(13)

(a) (b)

Figure 15. Effects of CG movement at Ma1 ¼ 2.0: (a)  vs. CGb; (b) CL/CD vs. CGb.

Figure 16. Optimization results at Ma1 ¼ 2.0 and H ¼ 300 m: (a) optimum configuration and (b) associated flow field.

presented. The bounds of design variables in opti- subsonic case, but the span is shorter. The associated
mization is shown in Table 6, in which the span lift-to-drag ratio is also reduced. More importantly,
takes values in the entire design space with s ¼ 1.8– the optimum CG is greatly moved backward
2.5 m, but the sweep angle is set from 5 to 55 and the (CGb ¼ 2.8 m). Figure 12 further demonstrates the
angle of attack is 0–2 based on a reasonable estima- effect of CG movement on the lift-to-drag ratio and
tion and pre-optimization. A medium-sweep wing has sweep angle. The CL/CD first increases with moving
better performance at a transonic speed through CG backward, then decreases, but the sweep angle
delaying occurrence of shock waves. The computa- monotonically increases owing to the static stability
tional domain is a hexahedron with 20 times body constraint. Relative to a fixed CG (CGb ¼ 2.3 m), the
length upstream of the model, 20 times body length movable CG can enhance the CL/CD with 30% in
downstream, and 20 times body length at the side. maximum. Figure 13 shows the optimal configuration
The grid cells are around 1.33 million that vary and associated pressure contour. The sub-surfaces of
slightly during the optimization due to the re-meshing the response surface are illustrated in Figure 14 with
of the volume meshes. A total of 7200 sampling points CL/CD against two design variables. The plots in
are selected for constructing response surface models. Figure 14(a) and (b) show a maximum about
The optimum configuration parameters are shown l ¼ 28 , but the optimum sweep is not the maximum
in Table 7, in which the optimum sweep angle owing to the constraints. CL/CD also basically mono-
becomes higher, compared with the preceding tonically increases up to the bounds of specified values
Gong and Ma 4995

Figure 17. Response surface at Ma1 ¼ 2.0 and H ¼ 300 m: (a) CL/CD vs  and s; (b) CL/CD vs.  and ; (c) CL/CD vs. s and ; (c) CL/CD
vs. f and .

Table 9. Optimization results at Ma1 ¼ 0.85.

Algorithm  (deg)  (deg) s (m) f (deg) CGb CL CL / CD CM @CM =@CL

Genetic 0.557 74.388 1.853 0.776 3.15 0.019 0.2564 0.000 0.122
Gradient 0.457 73.914 1.894 0.009 3.15 0.019 0.2558 0.000 0.146

with increasing the span and angle of attack, indicat- attack is set as 0.0–1.5 . A total of 3456 sampling
ing that the optimum values of design variables in points are calculated for constructing the response
Table 6 are determined by the constraints. surface.
In a supersonic simulation, the downstream dis-
turbances are unable to travel forwards across shock
Supersonic condition (Ma1 ¼ 2.0, H ¼ 300 m) waves, hence the upstream computational domain can
In this subsection, the optimum configuration for be smaller. The computational domain is still a hexa-
morphing aircraft at a supersonic and low-altitude hedron with five times body length upstream, 10 times
condition (Ma1 ¼ 2.0 and H ¼ 300 m) are presented. body length downstream, and 10 times body length at
Table 8 shows the range of design variables used, in the side. Approximately 1.22 million grid cells were
which the span s is still set as 1.8–2.5 m; since a high- employed that can vary slightly for various cases
sweep wing can have better performance at supersonic during the optimization. The results of the genetic
speeds, the sweep angle is set as 50 –80 ; since the and local gradient algorithm are shown in Table 9,
AOA is lower at a supersonic flight, the angle of in which the optimum sweep angle further increases
4996 Proc IMechE Part G: J Aerospace Engineering 233(13)

as expected, compared with the transonic case, but the Based on this optimization framework, a generic
span is similar. The lift-to-drag ratio is significantly morphing aircraft with variable sweep and span was
reduced owing to larger wave drag and smaller lift. investigated through the optimization process at sub-
The optimum CG is also further moved backward sonic, transonic and supersonic conditions, in which
(CGb ¼ 3.15 m). Figure 15 shows the effects of CG the morphing parameters consist of wing sweep angle
movement on CL/CD and l. The CL/CD first increases (0–80o), wing span (50% variation), and the target of
with moving the CG backward, but almost unchanged the optimization is to obtain maximum lift-to-drag
as CGb > 3.0 (Figure 15(a)). The sweep angle mono- ratios subject to lift, trim, and static stability con-
tonically increases owing to the static stability con- strains at each flight condition. The CG movement
straint with an exception of CGb ¼ 3.2 where l of the aircraft was also considered as a design variable
shows a reduction. Figure 16 shows the optimal con- in optimization. The results indicate that the CG has
figuration and associated pressure contour at the an important effect on the optimum configurations
supersonic case, in which shock waves around the air- obtained, and the aerodynamic performance will be
craft are visible around the fore-body and down- significantly enhanced if the CG can be moved back-
stream of aft-body. ward at transonic and supersonic cases. In the case of
The sub-surfaces of the response surface are illu- movable CG, the optimum sweep angles increase with
strated in Figure 17 to analyze the effects of design flight speeds increasing, and the optimum span at
variables on the lift-to-drag ratio with CL/CD vs. transonic and supersonic speeds is smaller than the
two design variables. The CL/CD exhibits a max- subsonic case.
imum at around l ¼ 65 (Figure 17(a) and (b)),
whereas the CL/CD monotonically grows up to the
upper bounds with increasing span and angle of Declaration of Conflicting Interests
attack, and therefore the optimum values for the The author(s) declared no potential conflicts of interest with
span and angle of attack in Table 9 is determined respect to the research, authorship, and/or publication of
by the lift constraint. this article.
Note that the UAV is designed for multimissions
with subsonic take-off at a low altitude, transonic Funding
cruise at a high altitude, and supersonic attack at a The author(s) received no financial support for the research,
low altitude. In the supersonic phase, none of current authorship, and/or publication of this article.
turbine engines are available to be able to propel the
vehicle at such a low altitude, and therefore a rocket ORCID iD
booster is necessarily attached to accelerate it to a Bao-Feng Ma [Link]
supersonic speed. In the mid-term future, an ongoing
turbine-based combined cycle (TBCC) engine will be References
probably adopted as one of the promising propulsion 1. Weisshaar TA. Morphing aircraft systems: historical per-
technologies. spectives and future challenges. J Aircraft 2013; 50:
337–353.
2. Vasista S, Tong L and Wong KC. Realization of morph-
Conclusions ing wings: a multidisciplinary challenge. J Aircraft 2012;
A robust and high-fidelity optimization framework 49: 11–28.
3. Ajaj RM, Beaverstock CS and Friswell MI. Morphing
for optimization design of morphing aircraft was
aircraft: the need for a new design philosophy. Aerosp Sci
established, by which the optimal aerodynamic con-
Technol 2016; 49: 154–166.
figuration corresponding to any flight condition can 4. Fincham JHS and Friswell MI. Aerodynamic optimisa-
be determined efficiently. The framework integrates tion of a camber morphing aerofoil. Aerosp Sci Technol
the existing codes, in which the configurations are 2015; 43: 245–255.
modeled in terms of individual components of air- 5. Gano SE, Renaud JE, Batill SM, et al. Shape optimiza-
craft, hence the large deformation of the aircraft can tion for conforming airfoils. In: 44th AIAA/ASME/
be easily dealt with during the optimization. The aero- ASCE/AHS structures, structural dynamics, and mater-
dynamic forces are computed by an inviscid Euler- ials conference, Norfolk, VA, USA, 7–10 April 2003,
based solver and friction/from drag estimation code. AIAA Paper 2003-1579.
A surrogate model is used for fitting the aerodynamic 6. Secanell M, Suleman A and Gamboa P. Design of a
morphing airfoil using aerodynamic shape optimization.
forces to reduce the computational cost in a global
AIAA J 2006; 44: 1550–1562.
optimization based on a generic algorithm. The
7. Namgoong H, Crossley WA and Lyrintzis AS.
method is effective and efficient, and is fully applicable Aerodynamic optimization of a morphing airfoil
to various morphing configurations with considerable using energy as an objective. AIAA J 2007; 45:
deformation and multiple variables. It also be able to 2113–2124.
obtain the optimal configurations of morphing air- 8. Gamboa P, Vale J, Lau FJP, et al. Optimization of a
craft for any flight condition within the range of morphing wing based on coupled aerodynamic and
morphing parameters. structural constraints. AIAA J 2009; 47: 2087–2104.
Gong and Ma 4997

9. Spearman ML. Aerodynamic research at NACA/ 25. Kenway GKW and Martins JRRA. Multipoint high-
NASA Langley related to the use of variable-sweep fidelity aerostructural optimization of a transport air-
wings. In: 50th AIAA aerospace sciences meeting includ- craft configuration. J Aircraft 2014; 51: 144–160.
ing the new horizons forum and aerospace exposition, 26. Biava M, Woodgate M and Barakos GN. Fully implicit
Nashville, TN, 9–12 January 2012, AIAA Paper 2012- discrete-adjoint methods for rotorcraft applications.
0956. AIAA J 2016; 54: 735–749.
10. Takahashi TT, Spall RJ, Turner DC, et al. A multi- 27. Chernukhin O and Zingg DW. Multimodality and
disciplinary assessment of morphing aircraft technology global optimization in aerodynamic design. AIAA J
applied to tactical cruise missile configurations. In: 45th 2013; 51: 1342–1354.
AIAA/ASME/ASCE/AHS/ASC structures, structural 28. Antunes AP and Azevedo JLF. Studies in aerodynamic
dynamics & materials conference, Palm Springs, CA, optimization based on genetic algorithms. J Aircraft
USA, 19–22 April 2004, AIAA Paper 2004-1725. 2014; 51: 1002–1012.
11. Bae J-S, Seigler TM and Inman DJ. Aerodynamic and 29. Poole DJ, Allen CB and Rendall TCS. High-fidelity
static aeroelastic characteristics of a variable-span aerodynamic shape optimization using efficient orthog-
morphing wing. J Aircraft 2005; 42: 528–534. onal modal design variables with a constrained global
12. Mestrinho J, Gamboa P and Santos P. Design opti- optimizer. Comput Fluids 2017; 143: 1–15.
mization of a variable-span morphing wing for a 30. Queipo NV, Haftka RT, Shyy W, et al. Surrogate-
small UAV. In: 52nd AIAA/ASME/ASCE/AHS/ASC based analysis and optimization. Prog Aerosp Sci
structures, structural dynamics and materials conference, 2005; 41: 1–28.
Denver, CO, 4–7 April 2011, AIAA Paper 2011-2025. 31. Forrester AIJ and Keane AJ. Recent advances in sur-
13. Ajaj RM, Friswell MI, Saavedra Flores EI, et al. Span rogate-based optimization. Prog Aerosp Sci 2009; 45:
morphing: a conceptual design study. In: 53rd AIAA/ 50–79.
ASME/ASCE/AHS/ASC structures, structural dynam- 32. Bevan RLT, Poole DJ, Allen CB, et al. Adaptive surro-
ics and materials conference, Honolulu, HI, USA, gate-based optimization of vortex generators for tiltro-
23–26 April 2012, AIAA Paper 2012-1510. tor geometry. J Aircraft 2017; 54: 1011–1024.
14. Woods BKS and Friswell MI. The adaptive aspect ratio 33. Han Z-H. Improving adjoint-based aerodynamic opti-
morphing wing: design concept and low fidelity skin mization via gradient-enhanced Kriging. In: 50th AIAA
optimization. Aerosp Sci Technol 2016; 42: 209–217. aerospace sciences meeting, Nashville, Tennessee,
15. Yang J, Cooper JE and Nangia RK. Optimization January 2012, AIAA Paper 2012-0670.
framework for design of morphing wings. In: 14th 34. Nemec M, Aftosmis MJ and Pulliam TH. CAD-based
AIAA aviation technology, integration, and operations aerodynamic design of complex configurations using a
conference, AIAA aviation forum, Atlanta, GA, 16–20 Cartesian method. In: 42nd AIAA aerospace sciences
June 2014, AIAA Paper 2014-2722. meeting, Reno, Nevada, 2004, AIAA Paper 2004-0113.
16. Ismail NI, Zulkifli AH, Abdullah MZ, et al. 35. McDonald R. Open VSP, [Link] (accessed
Optimization of aerodynamic efficiency for twist March 2017).
morphing MAV wing. Chinese J Aeronaut 2014; 27: 36. McDonald RA. Interactive reconstruction of 3D
475–487. models in the OpenVSP parametric geometry tool. In:
17. Hunsaker DF, Phillips WF and Joo JJ. Aerodynamic 53rd AIAA aerospace sciences meeting, AIAA SciTech
shape optimization of morphing wings at multiple flight Forum, Kissimmee, Florida 2015, AIAA 2015-1014.
conditions. In: 55th AIAA aerospace sciences meeting, 37. Aftosmis MJ. Cart3D Resource Website, [Link]
AIAA SciTech Forum, Grapevine, TX, USA, 9–13 [Link]/aftosmis/cart3d/[Link]
January 2017, AIAA Paper 2017-1420. (accessed March 2017).
18. Su WH, Swei SS-M and Zhu GG. Optimum wing shape 38. Ordaz I and Li W. Integration of off-track sonic boom
of highly flexible morphing aircraft for improved flight analysis in conceptual design of supersonic aircraft.
performance. J Aircraft 2016; 53: 1–12. J Aircraft 2014; 51: 23–28.
19. Prabhakar N, Prazenica R, Gudmundsson S, et al. 39. Ordaz I, Geiselhart KA and Fenbert JW. Conceptual
Transient dynamic analysis and control of a morphing design of low-boom aircraft with flight trim require-
UAV. In: AIAA SciTech forum, San Diego, CA, USA, ment. J Aircraft 2015; 52: 932–939.
4–8 January 2016, AIAA Paper 2016-0893. 40. Gur O, Mason WH and Schetz JA. Full configuration
20. Wickenheiser AM and Garcia E. Extended nonlinear drag estimation. J Aircraft 2010; 47: 110–116.
lifting-line method for aerodynamic modeling of recon- 41. The 2nd AIAA CFD Drag Prediction Workshop,
figurable aircraft. J Aircraft 2011; 48: 1812–1816. [Link]
21. Lyu Z and Martins JRRA. Aerodynamic shape opti- html (accessed March 2017).
mization of an adaptive morphing trailing-edge wing. 42. Sahu J and Heavey KR. Computations of supersonic
J Aircraft 2015; 52: 1951–1970. flow over a complex elliptical missile configuration. In:
22. Jameson A. Aerodynamic design via control theory. AIAA atmospheric flight mechanics conference, Chicago,
J Scient Comput 1988; 3: 233–260. IL, USA, 10–13 August 2009, AIAA Paper 2009-5714.
23. Hicken JE and Zingg DW. Induced-drag minimization 43. Langton R, Clark C, Hewitt M, et al. Aircraft fuel sys-
of nonplanar geometries based on the Euler equations. tems. New York: John Wiley & Sons, Ltd, 2009.
AIAA J 2010; 48: 2564–2575. 44. Açıkel HH and Serdar Genç M. Control of laminar
24. Gagnon H and Zingg DW. Euler-equation-based drag separation bubble over wind turbine airfoil using partial
minimization of unconventional aircraft configurations. flexibility on suction surface. Energy 2018; 165:
J Aircraft 2016; 53: 1361–1371. 176–190.